Enhancing process control education with the Control Station training simulator

Enhancing Process ControlEducation with the ControlStation Training SimulatorDOUG COOPER, DANIELLE DOUGHERTY

Department of Chemical Engineering, 191 Auditorium Road, Room 204, U-222, University of Connecticut,Storrs, Connecticut 06269-3222

Received 10 August 1999; accepted 25 October 1999

ABSTRACT: A process control training simulator can enhance learning by integratingthe theoretical abstraction of textbooks with the tactile nature of the lab and plant. Theprimary objective of a training simulator is education. It can motivate, help with visualiza-tion, and provide hands-on practice and experience. This article explores the use andbenefits of the Control Station training simulator for process control education. Examplespresented illustrate how the standard curriculum can be enhanced with a series of hands-onexercises and study projects. © 2000 John Wiley & Sons, Inc. Comput Appl Eng Educ 7: 203–212, 1999

Keywords: process control; training simulator; Control Station

INTRODUCTION

Practice in applying textbook control theory cangreatly benefit the learning process. Such practice ismotivating, promotes critical thinking, facilitates un-derstanding in the use and limitations of the theory,and helps prepare students for the challenges of theprofessional world. Too often, the application of text-book theory is limited to solving questions listed atthe end of the chapter. A typical question is to havethe student expand or extend a mathematical devel-opment presented in the book. Another is to providebits of data and then challenge the student to selectand employ a combination of formulas to obtain adesired result.

Unfortunately, even when cleverly crafted, theseone-dimensional challenges fall short of providingstudents the depth or breadth of practice required forlearning and comprehension. Thus, the Department of

Chemical Engineering at the University of Connecti-cut, like most around the world, supplements thetextbook with laboratory exercises. Hands-on labora-tory exercises are extremely important to learningbecause they help students make the intellectual tran-sition from theory to practice. The abstractions pre-sented in textbooks are literally brought to lifethrough the tactile nature of lab experience.

The reality of the laboratory, unfortunately, is thateach study can take many hours and even days toperform. Also, equipment failures and other problemsteach the important but not always appropriate lessonthat the real world can be uncertain. Thus, it can bedifficult to have the students explore more than a veryfew central concepts in the lab.

An alluring method for providing students with thesignificant hands-on practice critical to learning pro-cess control is with a training simulator that providesvirtual experience much the way airplane and powerplant simulators do in those fields. The proper tool canprovide students with a broad range of focused engi-neering applications of theory in an efficient, safe, and

Correspondence to D. J. Cooper ([email protected]).© 2000 John Wiley & Sons, Inc.

203

economical fashion. Such a simulator can work as aninstructional companion as it provides interactive casestudy challenges which track along with classroomlectures.

Process control is a subject area well suited to exploitthe benefits of a training simulator. Modern controlinstallations are computer based, so a video display is thenatural window through which the subject is practiced.With color graphic animation and interactive challenges,a training simulator can offer experiences which literallyrival those of the real world. These experiences can beobtained risk free and at minimal cost, enabling studentsto feel comfortable exploring nonstandard solutions attheir desk. If properly designed as a pedagogical toolwith case studies organized to present incremental chal-lenges, learning can be enormously enhanced for processcontrol.

A CHEMICAL PERSPECTIVE

Each discipline views process control from a differentperspective. To help orient the reader, consider thesetypical examples drawn from chemical process con-trol:

Figure 1 Gravity-drained tanks graphic display.

Figure 2 First-order plus dead time model fit of doublet test data.

204 COOPER AND DOUGHERTY

Process variables: temperature, pressure, pres-sure drop, level, flow, density, concentration

Final control elements: solenoid, valve, variablespeed pump or compressor, heater or cooler

Control algorithms: on/off, PID, cascade, ratio,feed forward, multivariable decouplers, model pre-dictive

Process Applications: reactors, separators, dis-tillation columns, heat exchangers, furnaces.

The chemical engineering perspective considers eachprocess to be one of a kind. Consequently, everycontrol system can appear unique in design, imple-mentation, and man–machine interface.

In addition, chemical processes are nonlinear andnonstationary, and have long time constants, signifi-cant dead time, and noisy measurement signals. Dis-turbances occur from numerous sources includingloop interaction from other controllers in the plant.

EXAMPLE LESSONS

The following lessons were drawn from the ControlStation [1] process control training simulator to illus-

trate the value such software provides the curriculum.We note that training simulators are distinguished inthis work from tools such as Matlab [2], which havea primary function of design, analysis, and simulation.The reader can download a free Control Station demoat www.engr.uconn.edu/control.

P-Only Controller Performance

The computer graphic display for the gravity-drainedtanks process, shown in Figure 1, is two vesselsstacked one above the other. Liquid drains freelythrough a hole in the bottom of each tank. The con-troller output signal manipulates the flow rate of liq-uid entering the top tank. The measured process vari-able is liquid level in the lower tank. The disturbancevariable is a secondary flow out of the lower tankfrom a positive displacement pump, so it is indepen-dent of liquid level except when the tank is empty.

Students begin their studies with this process be-cause its dynamic behavior is reasonably intuitive. Ifthey increase the liquid flow rate into the top tank, theliquid level rise in the tanks. If they decrease the flowrate, the level falls.

The traditional place to begin a course is with the

Figure 3 P-Only set point tracking results in offset.

CONTROL STATION TRAINING SIMULATOR 205

study of process dynamics. Students generate a steptest plot and compute by hand the first-order plus deadtime (FOPDT) model parameters: steady-state processgain,KP, overall time constant,tP, and apparent deadtime, uP. After they have gained mastery with handcalculations, they use tools that automate the model-fitting task so they can explore more practical tests. AControl Station fit of doublet test data is shown inFigure 2 for the gravity-drained tanks.

Students use their FOPDT model parameters intuning correlations to compute a P-Only controllergain,KC. Figure 3displays a simulator strip chart ofset point tracking performance for the gravity-drainedtanks under P-Only control. TheKC for the controlleris computed from the integral time weighted absoluteerror (ITAE) correlation using the FOPDT modelparameters from Figure 2.

Then comes the what-if studies. The investigationof Figure 4 explores howKC affects offset and damp-ing for set point tracking under P-Only control. Stu-dents also explore disturbance rejection under P-Onlycontrol. Is the best tuning for set point tracking thesame as for disturbance rejection? And how is besttuning defined?

PI Control and Nonlinear Behavior

The computer graphic for the countercurrent, shelland tube, lube oil cooler (a kind of heat exchanger) isshown in Figure 5. The controller output signal ma-nipulates the flow rate of cooling liquid on the shellside. The measured process variable is lube oil tem-perature exiting on the tube side.

Students learn an important lesson about processdynamics by studying the nonlinear character of thisprocess, as shown in Figure 6. The steady-state gainof the process clearly changes as operating levelchanges. Less obvious is that the time constant of theprocess also changes.

For processes that have such a nonlinear character,the performance of a controller will change as theprocess moves across operating levels. Figure 7 illus-trates this point. The exchanger is under PI controland as the set point is stepped to different operatinglevels, the nonlinear behavior of the process clearlyaffects set point tracking performance. Thus, studentslearn that a controller is designed for a specific ordesign level of operation. The best practice is tocollect dynamic test data as near as practical to thisdesign operating level.

Figure 4 P-Only performance changes asKC changes.


Figure 7 also shows that the heat exchanger has anegative steady-state gain. Students learn that a com-plete design includes specifying the sense of the con-troller (reverse versus direct acting). They learn thisconcept because if they enter it wrong, the controlleroutput will quickly drive the valve to either full openor full closed.

For what-if studies, students explore how PI con-troller tuning parameters interact and affect set pointtracking performance. Figure 8 shows a tuning mapthat they develop from an orderly tuning investiga-tion.

PID Control and Measurement Noise

Derivative action dampens oscillations because it re-sists rapid movement in the measured process vari-able. Students learn this by constructing Figure 9,which is a portion of a tuning map for derivativeaction. The center plot shows the set point trackingperformance of a PID controller tuned using the ITAEfor set point tracking correlation.

For all plots in Figure 9,KC andtI remain constantand there is no measurement noise. The plot to the leftshows how the oscillating nature of the responseincreases as derivative action is cut in half. The plot tothe right shows that when derivative action is too

Figure 6 Heat exchanger displays nonlinear behavior.

Figure 5 Heat exchanger graphic display.


Figure 8 PI controller tuning impacts performance.

Figure 7 Nonlinear behavior impacts performance.


large, it inhibits rapid movement in the measure pro-cess variable, causing the rise time and settling time tolengthen. When noise is added to the measured pro-cess variable, students learn that derivative actionamplifies it and reflects it in the controller outputsignal. Figure 10 illustrates this with a side-by-sidecomparison of a PI and PID controller.

Students also compare derivative on error to deriv-ative on measurement. Watching the derivative onerror “kick” after a set point step is a more memorableexperience than simply hearing about it. Derivativefiltering and four mode PID control are used in indus-try to address the problems resulting from measure-ment noise. Four-mode PID and filtering is scheduledto be added to Control Station in the near future.

Cascade, Feed Forward, and DisturbanceRejection

The jacketed reactor graphic, shown in Figure 11 forthe cascade case, is a continuously stirred tank reactorin which an irreversible exothermic reaction occurs.Residence time is constant in this perfectly mixedreactor, so the steady-state conversion from the reac-tor can be directly inferred from the temperature ofthe reactor product stream. To control reactor temper-ature, the vessel is enclosed with a jacket throughwhich a coolant passes.

The controller output manipulates the coolant flowrate through the jacket. The measured process vari-able is product exit stream temperature. If the exit

Figure 10 Measurement noise is amplified and reflected in controller output signal.

Figure 9 Derivative action affects oscillatory behavior.


stream temperature is too high, the controller in-creases the coolant jacket flow to cool down thereactor. The disturbance variable is the inlet temper-ature of coolant entering the cooling jacket.

The jacketed reactor can be run in three configu-rations: feedback control, feed forward with feedbacktrim, and cascade control as shown in Figure 11.

When the cooling jacket inlet temperature changes,the ability to remove heat changes and the controlsystem must compensate for this disturbance. Cascadeand feed forward are control strategies used for im-proved disturbance rejection. Cascade design involvesthe tuning of two controllers, as shown in Figure 11.Feed forward requires identification of an appropriateprocess and disturbance model.

The rejection of a change in the disturbance vari-able (jacket inlet temperature) for a single-loop PIcontroller is compared in Figure 12 with a PI withfeed-forward controller. The benefit of feed forward isclear for this process because for the same distur-bance, the process variable has a much smaller max-imum deviation and a faster settling time.

Students compare single loop, feed forward, andcascade control. They investigate tuning issues, whichPID modes to use in a cascade, the order of themodels needed for feed-forward design, plant–modelmismatch, dead time issues, and a host of other inter-esting challenges.

Control Loop Interaction and Decoupling

The distillation column graphic, shown in Figure 13,is a binary distillation column. The column has twomeasured process variables and two manipulated vari-ables. The reflux rate is used to control distillate

Figure 12 Benefits of feed-forward control.

Figure 11 Jacketed reactor under cascade control.


purity and the steam rate is used to control the purityof the bottoms stream.

Students use this process to explore the interac-tions that can occur in such multicontroller applica-tions. Control loop interaction occurs because whenthe distillate purity out of the top of the column is toolow, the top controller compensates by increasing the

flow of cold reflux into the column. This increasedreflux flow will indeed cause an increase in the dis-tillate purity. However, the additional cold reflux willwork its way down the column trays and eventuallybegin to cool the bottom of the column. This coolingcauses the purity of the bottoms stream to move offset point and produce a controller error.

The bottom controller compensates by increasingthe flow of steam into the reboiler. This produces anincrease in hot vapors traveling up the column, whicheventually causes the top of the column to begin toheat up. The result is that distillate purity again be-comes too low. In response, the top controller com-pensates by again increasing the flow of cold refluxinto the column.

This controller “fight” is shown on the left side ofFigure 14. The upper trace shows the distillate com-position responding to a step set point change. Con-troller interaction causes the bottoms composition,shown in the lower trace, to react unfavorably.

Decouplers are feed-forward elements where themeasured disturbance is the controller output signal ofanother loop on the process. Two decouplers are re-quired to compensate for loop interaction, one foreach controller. Like a feed-forward element, eachdecoupler requires identification of a process and dis-turbance model. The right side of Figure 14 showsthat with decouplers in place, this loop interaction isdramatically reduced.

Figure 14 Distillation column shows loop interaction..

Figure 13 Distillation column graphic display.


Students explore different controller modes, looptunings, model structures, and many other design is-sues. With two controllers and four models for com-plete decoupling, students also learn how importantbookkeeping is to the control designer.

CONCLUSION

We have presented some examples of the lessons andchallenges a training simulator can provide. Spaceprohibits presentation of other studies available inControl Station, including the control of integratingprocesses, the use of the Smith predictor model pre-dictive controller, and a host of process identificationmethods and procedures.

We stress that we do not believe a training simu-lator is better than or a replacement for real labexperiences. In fact, we believe that hands-on studies

with actual equipment are fundamental to the learningprocess. We are of the opinion, however, that a propertraining simulator can provide students with a broadrange of meaningful experiences in a safe and effi-cient fashion. These experiences can be obtained riskfree and at minimal cost, enabling students to feelcomfortable exploring nonstandard solutions at theirdesk. If properly designed, a training simulator canbridge the gap between textbook and laboratory, en-abling significantly enhanced learning for processcontrol theory and practice.

REFERENCES

[1] D. J. Cooper and M. Sinha, Control Station educa-tional software, CACHE News 44 (1997), 14–21.

[2] D. Hanselman and B. Littlefield, The student edition ofMatlab, Prentice-Hall, Englewood Cliffs, NJ, 1995.

BIOGRAPHIES

Doug Cooper is a professor of chemicalengineering at the University of Connecticut.He has been teaching and directing researchin process control for more than 15 years.His research focuses on developing controlmethods that are both reliable and easy forpractitioners to use. He has studied the con-trol of fluidized bed incineration, heat ex-change, distillation, injection molding, surge

tanks, and catalytic reactors. Dr. Cooper’s work experience inindustry sparked his interest in teaching process control from areal-world perspective. This interest ultimately led him to developthe popular Control Station process control training simulator de-scribed here.

Danielle Dougherty was in born in Phila-delphia, Pennsylvania, on September 15,1975. She received a BS in chemical engi-neering from Widener University in 1997.She is currently working toward a PhD de-gree in chemical engineering under the di-rection of Doug Cooper at the University ofConnecticut. Her current research interestsincluded model predictive control and non-

linear adaptive control. Her thesis is on multivariable adaptivemodel predictive control.


Enhancing process control education with the Control Station training simulator

Documents

Transcript of Enhancing process control education with the Control Station training simulator